OpenAI Unveils Secure MCP Tunnel

This is a solution to address the security risk of exposing servers in a company's private environment to the outside when using the Model Context Protocol (MCP).

The core of Secure MCP Tunnel lies in reversing the direction of the connection. While the existing method was an inbound approach coming from outside to inside, this technology takes a structure in which the client in the private environment first establishes an outbound HTTPS connection toward OpenAI.

The main features of this approach are as follows:

• Maintaining security boundaries: Servers can be kept under existing network controls without needing to open public endpoints or extend a VPN.
• Minimizing third-party dependencies: It directly leverages the OpenAI-hosted environment without adding a separate vendor such as ngrok or Cloudflare Tunnel.
• Efficient connectivity: It supports MCP's streaming and notification features, and can be operated via a small, verifiable open-source client.

As a result, enterprise users can keep their private tools and data secure while easily connecting to hosted AI products such as ChatGPT or Codex.